Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=gcec20

Download by: [NUS National University of Singapore] Date: 14 January 2016, At: 17:27

Chemical Engineering Communications

ISSN: 0098-6445 (Print) 1563-5201 (Online) Journal homepage: http://www.tandfonline.com/loi/gcec20

PLANTWIDE CONTROL STUDY OF A VINYL ACETATEMONOMER PROCESS DESIGN

Donald G. Olsen , William Y. Svrcek & Brent R. Young

To cite this article: Donald G. Olsen , William Y. Svrcek & Brent R. Young (2005) PLANTWIDECONTROL STUDY OF A VINYL ACETATE MONOMER PROCESS DESIGN, Chemical EngineeringCommunications, 192:10, 1243-1257, DOI: 10.1080/009864490515711

To link to this article: http://dx.doi.org/10.1080/009864490515711

Published online: 25 Jan 2007.

Submit your article to this journal

Article views: 220

View related articles

Citing articles: 9 View citing articles

Plantwide Control Study of a Vinyl AcetateMonomer Process Design

DONALD G. OLSEN, WILLIAM Y. SVRCEK,AND BRENT R. YOUNG

Department of Chemical and Petroleum Engineering, University of Calgary,Calgary, Alberta, Canada

A simulation of a vinyl acetate monomer (VAM) process design was developed andcompared with the work of Luyben and Tyreus (1998). Two incremental changeswere made to the two main control substructures. Specifically, the two schemes focuson improving the liquid inventory system control and controllability of the azeotropicdistillation column. The level control strategy was tested and found to produce a fas-ter response with less oscillatory behavior. Two alternative control techniques for theazeotropic distillation column were tested, a feed-forward model predictive control-ler and a static feed-forward ratio controller. The model predictive controller resultsillustrated the large difference between the water composition analyzer sample timeand the controller step size. The static feed-forward ratio controller showed excellentdisturbance rejection of large feed flow variations to the azeotropic distillationcolumn. Simulation results are presented to illustrate the effectiveness of the newcontrol strategies.

Keywords Vinyl acetate monomer; Simulation; Model predictive control; Ratiocontrol

Introduction

Plantwide control system design and optimization has for many years been a recog-nized challenge in the research community (Mellichamp, 1993), and it continues tobe a major topic of interest, with the focus on new methods for creating the mosteffective control structures. However, the task is a challenge due to the large numberof variables involved and the resulting number of alternatives in plantwide controlsynthesis. Present-day techniques to overcome this difficulty range from thedesigners experience to formal sets of heuristics (Zheng et al., 1999, Luyben et al.,1997), to optimization with both steady-state and dynamic simulation (McAvoy,1999).

Today it is recognized that one key tool to be used in designing more effectivecontrol structures is dynamic simulation. With the aid of simulation, both researchand industrial practitioners can test their ideas and gain insight into process behaviorthat would not normally be intuitive given the complexity of an entire process design.Unfortunately for the research world, much plantwide information is proprietaryand not available in open literature (Mellichamp, 1993). However, in recent years

Address correspondence to Brent R. Young, Department of Chemical and PetroleumEngineering, University of Calgary, 2500 University Dr., NW, Calgary, Alberta T2N 1N4,Canada. E-mail: byoung@ucalgary.ca

Chem. Eng. Comm., 192:12431257, 2005Copyright # Taylor & Francis Inc.ISSN: 0098-6445 print/1563-5201 onlineDOI: 10.1080/009864490515711

1243

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

several good case studies have been published that allow testing of new control ideason the level of complexity seen in a typical industrial chemical manufacturing plant.

One such study that has been presented by Luyben and Tyreus (1998) providesdetailed process and rating information for the manufacture of vinyl acetate mono-mer (VAM). In addition, the authors propose and test an entire control structureusing their nine-step approach to plantwide control design (Luyben et al., 1999).This study uses their work as a base case for testing several proposed changes to theirdecentralized control strategy. A simple incremental approach has been applied thatlooks for ways of improving the dynamic controllability of the base design. Theresults and conclusions of a new reboiler level strategy, use of a feed-forward modelpredictive control (MPC), and a simple static ratio controller are presented.

Background Information

Process Description and Simulation

In the process to be studied, the production of VAM is based upon the common gasphase catalyzed reaction of ethylene, acetic acid, and oxygen. In addition, a domi-nant CO2 producing side reaction also occurs, consuming the key ethylene feedstock.It is assumed for this study that these two chemical reactions are the only ones thatoccur and are shown as follows:

C2H4 CH3COOH 0:5O2 , CH2CHOCOCH3 H2O 1C2H4 3O2 , 2CO2 2H2O 2

Figure 1 shows a schematic of the process. Only the central production units areincluded; the feed storage and product purification stages are not included as theyare not germane to this study. The major unit operations include: a vaporizer,

Figure 1. Overall process flow schematic.

1244 D. G. Olsen et al.

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

a gas phase plug flow reactor, a heat exchanger, a vapor=liquid separator, an absor-ber column, and an azeotopic distillation column. There are two major recycle loops,one for gas reactant recovery and the other for liquid reactant recovery. In addition,there is a secondary liquid recycle stream to the absorber for the recovery of liquidcarryover in the gas stream. Both liquid recycle loops are of particular interest asthey introduce a dynamic that is dominated by the effects of lag.

The process simulation has been based on the design data given in Luyben andTyreus (1998) and accomplished using the commercial process simulator HYSYSTM.Standard unit operations were used for the entire flow sheet simulation. However,ActiveXTM extensions were written to model the complex structures of the reactionrate expressions defined in Luyben et al. (1999), representing the kinetic behavior ofreactions (1) and (2). The HYSYS models developed in this article were provided toHyprotech Ltd.

To provide a more realistic basis for testing, the effects of sensor noise, deadtime, and sample time have also been included. Specifically, process noise of 1%of each process variable range has been added to process and measured variables.In addition, a filter was applied to each measured variable. Finally, the effects ofsample time and dead time resulting from the use of a water analyzer at the bottomof the distillation column have been assumed. A value of five minutes has been selec-ted for this sample time.

The UNIQUAC (Abrams and Prausnitz, 1975) thermodynamic model was selec-ted to model the VAM=water=acetic acid ternary mixture. The thermodynamicmodel interaction parameters were taken from low-pressure vapor-liquid equilib-rium data given in the DECHEMA data series (Gmehling and Onken, 1977). TheUNIFAC local linear estimator (LLE) estimation procedure (Frendenslund et al.,1975) was used to derive the VAM=water binary interaction parameters due to thelack of quantitative experimental LLE data for this binary. The liquid phase fugaci-ties of the noncondensable components were derived using Henrys Law coefficients.However, the ethane parameters used were the same as water in order to bettermatch material balance information given in Luyben and Tyreus (1998). The vaporphase fugacities were calculated using the ideal gas law.

Control Objectives

In order to establish an understanding for the plantwide control system, a clearstatement of the control objectives must be stated. For this VAM process designthe control objectives, as outlined in Luyben et al. (1999), are:

1) Set the production of VAM while minimizing yield losses to carbon dioxideproduction.

2) Absorber must recover as much of the VAM, water, and acetic acid from the gasrecycle loop to prevent losses to the CO2 removal system.

3) Provide control during production turndown.4) Oxygen concentration in the gas recycle loop must remain outside the explosivity

range for ethylene.5) Minimize losses of acetic acid to the overhead of the azeotropic distillation

column.6) Minimize VAM concentration in the bottoms stream of the azeotropic distillation

column to below 100 ppm.

Plantwide Control Study 1245

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

For this study, focus has been on objectives 5 and 6 above. The remaining objec-tives are handled by the base case plantwide control scheme discussed in the next sec-tion. For proper operation of the process several specific control requirementsbecome apparent from these objectives. These specific control requirements are listedbelow and have been adapted from Luyben et al. (1999).

1) Provide set point tracking of key process variables (reactor outlet temperature,azeotropic distillation column bottoms waster composition, carbon dioxide andethane concentration in the gas recycle stream).

2) Provide control with a water composition analyzer sample time of five minutesand demonstrate controllability under the influence of process noise.

3) Provide disturbance rejection due to a five-minute feed flow shutoff of the azeo-tropic distillation column feed pump.

4) Provide stable inventory control.

The focus of test work in this study has been on meeting these control require-ments and searching for methods to improve their performance.

Base Case Plantwide Control

The derivation and the performance of the base plantwide control has been welldocumented in Luyben et al. (1999). The simulation used in this study has been com-pared against the results of this reference and found to agree well with the documen-ted behavior. Figure 2 shows the base case control structure.

Of particular interest are the reboiler and the water composition control schemesof the azeotropic distillation column. In the base scheme, both the wash acid recyclestream and the total acetic acid feed flow to the vaporizer are on flow control. There-fore, changes in the reboiler level are controlled via the change in fresh feed flow rateof acetic acid. This pairing is justified for several reasons: the first is the need toprevent instability or snowballing of the liquid recycle loop by having the recycle

Figure 2. Base plantwide control strategy (Luyben et al., 1997).

1246 D. G. Olsen et al.

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

stream on flow control; second, putting the total acetic acid flow to the vaporizer onflow control ensures proper acetic acid concentration to the reactor; third, given thatthe reboiler level can be considered to be an indication of the plant acetic acid inven-tory, it is best to pair fresh feed flow with it. With this approach, the vaporizer levelis free to be paired directly to the heat input to the vaporizer.

The azeoptropic distillation column composition control is managed using twocontrol loops: a bottoms water-reflux loop and a stage 14 temperature-reboiler dutyloop. Interestingly, the water middle boiler and not the overhead acetic acid concen-tration is controlled. This approach ensures the overall plant water componentinventory is regulated. The stage 14 tray temperature loop is key to maintainingthe temperature profile of the column, thus ensuring the VAM concentration is lessthan 100 ppm in the bottoms stream.

The control loops were tuned using an auto tuning variation relay technique(Astrom and Hagglund, 1984). All pressure and level loops were set on proportionalcontrol only with a high gain for tight pressure regulation and averaging control forthe levels.

In this work, incremental changes were made to the base design by evaluatingtwo main control substructures. A new reboiler level control strategy was implemen-ted along with two alternative control arrangements for the azeotropic distillationcolumn. The first is the implementation of an MPC centralized controller and thesecond is a static feed-forward ratio controller. The results of each are comparedto the base case performance in the following sections.

Base Case Control Performance

Two tests were selected for study based on the specific control requirements men-tioned above. There was a step change from 0.09 to 0.18 mole fraction water com-position in the distillation column bottoms and an interruption in distillationcolumn feed flow for five minutes. Figure 3 shows the results for the first test.The response proved to be underdamped, resulting in an overshoot of the set pointand oscillatory behavior in liquid inventory levels and flows. The cause of this oscil-latory behavior is attributed primarily to the structure of the base case reboiler levelcontrol strategy. With the fresh acetic acid feed introduced immediately upstream ofthe vaporizer, a considerable amount of dead time is introduced into the system. As aresult, the interaction lag between the reboiler level and the fresh acetic acid feedadversly affects all other flow rate and composition controls. However, excellentset point tracking was attained for the stage 14 temperature controller. Throughthe tight temperature control, the VAM composition in the column bottoms streamremained below the specified 100 ppm constraint, reaching a maximum of only10 ppm.

Figure 4 illustrates the ability of the base case to reject the five minute feed flowinterruption disturbance. Due to the short time duration, the capacity controls werenot significantly affected. However, composition and temperature effects were sig-nificant. Despite the initial large variations in these variables, the process wasbrought under control and stabilized at the appropriate set points after the pumpwas restarted. The water composition response took approximately seven hours toreach steady state whereas the stage 14 temperature took only one hour. Finally,the VAM concentration in the bottoms reached a maximum of 40 ppm, remaining

Plantwide Control Study 1247

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

Figure 4. 5-minute shut-off of azeotropic distillation column feed pump base case.

Figure 3. Step test of WATER composition in azeotropic distillation column bottoms from0.09 to 0.18 mole fraction base case.

1248 D. G. Olsen et al.

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

below the 100 ppm limit. Again this result is attributed to the fast temperaturecontrol loop dynamics.

Plant Inventory Control Structure

Design

The design of the new reboiler level control structure is shown in Figure 5. In thisarrangement the acetic acid feed flow is paired with the vaporizer level and the dis-tillation column reboiler level is paired with the bottoms flow. The decision to usethis control structure was for two reasons: to minimize the dead time introduced intothe process by the base case design as discussed above and to maintain a constanttemperature of the vaporizer overhead gas stream. It is concluded that any undesir-able propagation of disturbances in the liquid recycle loop would be held in check.That is, the acetic acid feed flow is being used to directly control the level of thevaporizer, minimizing dead time in the system. Second, the vaporization rate remainsfixed by using a temperature controller on the vaporizer outlet. Therefore, the liquidrecycle is essentially under flow control preventing any snowball effects due to dis-turbances in the liquid inventory system. Pairing of reboiler level with bottoms flowresults in rapid rejection of any fluctuations in level and passes the flow rate changesonto the vaporizer level control.

However, two implementation issues should be noted. First, extra precautionwould have to be taken to ensure that the temperature sensor is accurate, as a faultysensor can result in little or no vaporization. Therefore, operations would be respon-sible for ensuring the integrity of this temperature sensor. In addition, acetic acidinventory is now regulated within the vaporizer given the pairing of acetic acid freshfeed with vaporizer level. Therefore, depending on the dynamics, the process design

Figure 5. New reboiler level control structure.

Plantwide Control Study 1249

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

may need to be changed to incorporate more capacity in the vaporizer or utilize anintermediate tank in the acetic acid recycle loop.

Results

To test the new configuration, the same step test of WATER bottoms compositionfrom 0.09 to 0.18 mole fraction was performed. The results shown in Figure 6 verifystable capacity and composition control of the key process variable. In comparisonto Figure 3, better performance was achieved, as less time was required to stabilizethe variables at their set points. The conclusion is that given the elimination ofadditional dead time in the reboiler level control strategy, less oscillatory behavioroccurred in the process, resulting in better performance.

Distillation Column Composition Control

Feed-Forward MPC

The first modification to the composition control substructure in the azeotropic dis-tillation column was to implement a model predictive controller (MPC). The reasonthat MPC was chosen is that it has become arguably the most widely implementedadvanced process control strategy in industry (Seborg, 1994). The controller wasapplied to the 2 2 composition control loop consisting of the water compositionin the bottoms stream and the tray 14 stage temperature. Due to the nature of theinvestigation in implementing an MPC, we decided to minimize the number of vari-ables at two. We selected the composition control loops in light of their importancefor product purity and overall operation stability, these loops being key to quality

Figure 6. Step test of WATER composition in azeotropic distillation column bottoms from0.09 to 0.18 mole fraction new reboiler level control.

1250 D. G. Olsen et al.

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

control for this distillation column (Olsen et al., 2002) and assuming the work ofLuyben et al. (McAvoy, 1999; Luyben et al., 1999) in developing the regulatory con-trol structure was optimal in some sense. Note that is also important to consider therestrictions of the manipulated variables in the development of an MPC. However,no data related to restrictions on manipulated variables were given in the originalreference. As a result this was considered in the following manner. All valves weresimply sized using standard control valve Cv sizing algorithms. The values werechecked to ensure saturation did not occur. The details of the controller and theresults are presented in the following sections.

Design DetailsThe MPC algorithm used in this study is the well-known dynamic matrix controller(DMC) formulation (Cutler and Ramaker, 1979). For this study, a feed-forwardcomponent was included in the overall calculation to account for feed flow distur-bances. Therefore, the full prediction horizon can be calculated as:

Y Ypast YFF A delU b 3where Ypast accounts for the effects of past moves in the manipulated variable, A isthe dynamic matrix of step coefficients, delU is the future moves of manipulatedvariables, and b accounts for model error. To account for measured disturbancevariables, a feed-forward component YFF is applied to the truncated step responsemodel based on the approximation of constant future disturbances (Hokansonand Gerstle, 1992). The step response models were easily obtained by introducingpositive changes in the manipulated variables. Given all this information, futuredelU moves are calculated using standard least squares to find the minimum errorfrom set point with only the first move actually being implemented.

To avoid the complexity of the open-loop behavior as discussed in Olsen et al.,(2002), a partial closed-loop strategy has been used. As a result, the temperaturecontroller set point is controlled directly as shown in Figure 7. The MPC controller

Figure 7. High-level MPC design showing relation between manipulated and controlledvariables.

Plantwide Control Study 1251

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

calculations were implemented in a custom computer program linked together withHYSYSTM and MatlabTM (van der Lee et al., 2001).

Due to the aggressive settings of the stage 14 temperature control loop, theywere detuned to provide a more critically damped response. This action was neces-sary in order to obtain an adequate step response model to allow stable performanceof the controller. Using the detuned control loop proved beneficial in two aspects:the oscillatory behavior of the underdamped response was removed and the differ-ence between the two composition controller time constants was reduced to anadequate level. A 30-second step size was selected based on the time constant ofthe detuned temperature control loop. Also, the step response model horizon wasselected to be 200 step coefficients. Tuning of the MPC controller used the methodin Maurth et al. (1988) whereby a control horizon of 1 was set and the predictionhorizon adjusted as a tuning parameter. Finally, the weighting and move suppressionmatrix were set to the identity matrix to provide equal weighting to each controlvariable and maintain conservative movements in the controller actions.

ResultsPerformance of the MPC controller is shown in Figures 8 and 9. Performance provedunacceptable for a set point change in water composition and acceptable for afive-minute feed flow interruption. The VAM concentration in the bottoms streamviolated the constraint of 100 ppm, as shown in Figure 8. In the case of feed flowdisturbance the control scheme was able to meet the control objectives. However,the temperature on stage 14 was poorly regulated, resulting in a large drop in tem-perature. Due to this poor temperature regulation, it may have been possible for theVAM concentration to rise above the 100 ppm constraint. Even though in this case it

Figure 8. Step test of WATER composition in azeotropic distillation column bottoms from0.09 to 0.18 mole fraction feed-forward MPC.

1252 D. G. Olsen et al.

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

did not, there is potential for the VAM to rise to unacceptable levels given the lowtemperatures on stage 14 required by the controller.

The cause of the poor performance is attributed to the difference in sample timeof the water composition controller relative to the controller sample time (five min-utes versus 30 s). As a result, the temperature response is too aggressive given thedelay in water composition measurement, resulting in too low a stage 14 temperatureand as shown in Figure 8 too high a VAM concentration in the column bottomsstream. It is postulated that a continuous measurement via an inferred water compo-sition such as a Kalman filter, neural network, or other soft sensors could allowadequate performance of the MPC controller in this situation (Quek et al., 2000).

Static Ratio Control

Design DetailsThe final test involved using a static ratio control scheme within the azeotropic dis-tillation column composition control substructure (Ryskamp, 1980). Ratio controlof this type is an example of a more advanced regulatory control strategy than thebase regulatory control structure. However, we are assuming the work of Luybenet al. (McAvoy, 1999; Luyben et al., 1999) in developing the regulatory control struc-ture was optimal, in some sense. We did not try to further improve their regulatorystructure and instead focused on classical advanced control techniques. In thisarrangement, a mass ratio of reflux flow to feed flow (L=F) is utilized to minimizethe effects on bottoms water composition due to fluctuations in the feed flow ratevariable. It is reasoned that effects of feed flow would be a frequent process disturb-ance, and therefore accounting for the effects in a feed forward manner would be asimple yet effective control method. The ratio of reboiler duty to feed-flow was not

Figure 9. 5-minute shut-off of azeotropic distillation column feed pump feed-forward MPC.

Plantwide Control Study 1253

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

implemented, as it was concluded that the PI stage 14 temperature contoller wasaggressive enough to reject temperature disturbances and would not provide signifi-cant control benefits. Figure 10 shows the control scheme using L=F as the newcontrol variable.

Results

The results for the two test cases are shown in Figures 11 and 12. In the first test, theperformance of the ratio scheme in changing the water composition proved to be thesame as in the base case. The cause is attributed to the dominant effects of reboilervaporization rate on water composition in comparison to the resulting small feedflow variation. Therefore, it is concluded there is no advantage to using the ratioscheme for water composition changes. However, the second test showed improvedperformance at minimizing the effects of a significant feed flow interruption. Inparticular, both the changes in water composition and stage 14 temperature variablesdeviated less from their set points and stabilized more quickly. The control schemewas able to immediately start to account for the drop in feed flow, providing the con-trol action before the composition controller measured any changes. Through theuse of the static L=F ratio, the composition controller was required to make lessaggressive reflux flow adjustments. It is concluded that a static L=F ratio controlleris a good column controller design option for feed flow disturbance rejection.

Figure 10. Static feed-forward control scheme.

1254 D. G. Olsen et al.

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

Conclusions

Dynamic modeling of a plantwide regulatory control system for a VAM process designhas been completed and compared with the work of Luyben and Tyreus (1998).

Figure 12. 5-minute shutoff of azeotropic distillation column feed pump static feed-forwardratio control.

Figure 11. Step test of WATER composition in azeotropic distillation column bottoms from0.09 to 0.18 mole fraction static feed-forward ratio control.

Plantwide Control Study 1255

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

Incremental changes were made to two main control substructures, the liquidinventory system and the 2 2 composition control of the azeotropic distillationcolumn. A reboiler level control strategy was tested and found to produce a fasterresponse with less oscillatory behavior. Two alternative control techniques for theazeotropic distillation column tested included a feed-forward model predictive con-troller and static feed-forward ratio controller. The model predictive controllerresults were poor due to the large difference between the WATER compositionanalyzer sample time and the controller step size. As a result, it was concluded thatoptions for an inferred composition analyzer would be needed to effectivelyimplement a model predictive controller for the VAM azeotropic distillation column.Finally, the static feed-forward ratio controller was implemented to maintain a con-stant reflux to feed flow ratio. The results showed excellent disturbance rejection oflarge feed flow variations to the azeoptropic distillation column. This study representsa thorough simulation study and is valuable as a reference for future work.

Acknowledgments

This article is an extended version of an original presentation made at the 2001Annual AIChE meeting, Reno, Nevada, November 49, 2001.

References

Abrams, D. S. and Prausnitz, J. M. (1975). Statistical thermodynamics of liquid mixtures: Anew expression for the excess Gibbs energy of partly or completely miscible systems,AICHE J., 21, 116128.

Astrom, K. L. and Hagglund, T. (1984). Automatic tuning of simple regulators with specifica-tions of phase and amplitude margins, Automatica, 20, 645651.

Cutler, C. R. and Ramaker, B. L. (1979). Dynamic matrix control: A computer control algor-ithm, paper presented at AIChE Meeting, Houston, Tex.

Frendenslund, A., Jones, J. M., and Prausnitz, J. M. (1975). Group contribution estimationsof activity coefficients in non-ideal liquid mixtures, AICHE J., 21, 10861098.

Gmehling, J. and Onken, U. (1977). Vapour-Liquid Equilibrium Data Collection, Vol. 1,DECHEMA, Frankfurt.

Hokanson, D. A. and Gerstle, J. G. (1992). Dynamic matrix control multivariable controllers,in Practical Distillation Control, ed. W. L. Luyben, 248271, Van Nostrand Reinhold,New York.

Luyben, M. L. and Tyreus, B. D. (1998). An industrial design=control study for the vinyl acet-ate monomer process, Comput. Chem. Eng., 22, 867877.

Luyben, M. L., Tyreus, B. D., and Luyben, W. L. (1997). Plantwide control design procedure,AIChE J., 43, 3161.

Luyben, W. L., Tyreus, B. D., and Luyben, M. L. (1999). Plantwide Process Control,McGraw-Hill, New York.

McAvoy, T. J. (1999). Synthesis of plantwide control systems using optimization, Ind. Eng.Chem. Res., 38. 29842994.

Maurth, P. R., Mellichamp, D. A., and Seborg, D. E. (1988). Predictive controller design forsingle-input=single-output (SISO) systems, Ind. Eng. Chem. Res., 27, 956963.

Mellichamp, D. A. (1993). Foreword Industrial challenge problems in process control,Comput. Chem. Eng., 17, vix.

Olsen, D. G., Young, B. R., and Svrcek, W. Y. (2002). A study in advanced control appli-cation to an azeotropic distillation column within a vinyl acetate monomer processdesign, Dev. Chem. Eng. Miner. Process., 10, 4760.

1256 D. G. Olsen et al.

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016

Quek, C. J., Balasubramanian, R., and Rangaiah, G. P. (2000). Consider using soft analyzersto improve SRU control Researchers use neural networks and regression methods tobuild models that accurately predict H2S and SO2 in tail gas, Hydrocarbon Process.,79(1), 101106.

Ryskamp, C. J. (1980). New strategy improves dual composition column control, Hydro-carbon Process., 60(6), 5159.

Seborg, D. E. (1994). A perspective on advanced strategies for process control,Model. Identif.Control, 15, 179189.

van der Lee, J. H., Olsen, D. G, Young, B. R., and Svrcek, W. Y. (2001). An integrated, realtime computing environment for advanced process control development, Chem. Eng.Educ., 35(3), 172179.

Zheng, A., Mahajanam, R. V., and Douglas, J. M. (1999). Hierarchical procedure for plant-wide control system synthesis, AIChE J., 45, 12551265.

Plantwide Control Study 1257

Dow

nloa

ded

by [N

US N

ation

al Un

iversi

ty of

Sing

apor

e] at

17:27

14 Ja

nuary

2016